Btrfs devel - Oct 2007 - [patch]Fix ENOTEMPTY check in btrfs_rmdir

Hello,

The ENOTEMPTY check in btrfs_rmdir isn't reliable. It's possible that
the backward search finds . or .. at first, then some other directory
entry. In that case,  btrfs_rmdir delete . or .. improperly.  The
patch also fix a fs_mutex unlock issue in  btrfs_rmdir by the way.

Regards
YZ

--
diff -r 9bf00d0c038f inode.c
--- a/inode.c	Fri Oct 19 14:01:21 2007 -0400
+++ b/inode.c	Thu Oct 25 19:29:48 2007 +0800
@@ -448,6 +448,9 @@ static int btrfs_rmdir(struct inode *dir
 	char *goodnames = "..";
 	unsigned long nr;

+	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
+		return -ENOTEMPTY;
+
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 	mutex_lock(&root->fs_info->fs_mutex);
@@ -501,9 +504,9 @@ out:
 out:
 	btrfs_release_path(root, path);
 	btrfs_free_path(path);
-	mutex_unlock(&root->fs_info->fs_mutex);
 	nr = trans->blocks_used;
 	ret = btrfs_end_transaction(trans, root);
+	mutex_unlock(&root->fs_info->fs_mutex);
 	btrfs_btree_balance_dirty(root, nr);
 	if (ret && !err)
 		err = ret;

On Thu, 25 Oct 2007 19:30:47 +0800
"Yan Zheng" <yanzheng@21cn.com> wrote:
> Hello,
> 
> The ENOTEMPTY check in btrfs_rmdir isn't reliable. It's possible
that
> the backward search finds . or .. at first, then some other directory
> entry. In that case,  btrfs_rmdir delete . or .. improperly.  The
> patch also fix a fs_mutex unlock issue in  btrfs_rmdir by the way.
Thanks, you're right.  The original code was before the inode number
index, when the backward search was correct.

-chris